Accurate land area estimates form the foundation of reliable agricultural statistics, providing critical inputs for estimating crop production, calculating yield, and informing key agricultural policy decisions. Traditional ground surveys where these data are collected, however, face significant resource constraints, especially in the Pacific Islands, where environmental conditions, physical distance between islands, and natural barriers can make data collection cost-prohibitive.
In 2022, the Asian Development Bank (ADB), in collaboration with the Cook Islands Ministry of Agriculture (MOA), conducted a post-enumeration survey on the island of Rarotonga to validate agricultural area estimates using global positioning systems (GPS) reported in the 2021 Cook Islands Census of Agriculture. The study revealed that farmers generally underestimated their agricultural land areas by approximately 13% relative to objective GPS measurements, with larger plots (>2,000 m²) more likely underestimated and smaller plots slightly overestimated (ADB, 2024) 1. This discrepancy highlights the significance of non-sampling errors, specifically measurement error, that can systematically bias national agricultural production estimates.
In a follow-up activity conducted in 2023, the ADB team with the MOA employed Unmanned Aerial Vehicles (UAVs) to capture high-resolution imagery and measure agricultural land areas, aiming to reduce resource intensity of field surveys in environmentally challenging terrain. While initial resource constraints may be higher, high-resolution imagery captured by UAV enabled precise plot boundary delineation and area measurements. UAVs were found to be particularly suitable for the Pacific context, where frequent cloud cover, fragmented land plots, and difficult terrain require flexible, localized, and detailed data collection methods.
Drawing on the 2023 Cook Islands UAV survey of a census enumeration area on Rarotonga, this chapter has two aims: (i) to provide a reproducible, end-to-end workflow in R — from loading orthophotos and vector boundaries, harmonizing CRS, and visualizing imagery to computing plot-level zonal statistics — and (ii) to demonstrate, through a worked case study, how very high-resolution data (≈1.6-cm GSD) enables agricultural statistics that are difficult or impossible with 10–30 m satellite sensors. Using RGB and multispectral imagery, we quantify plot areas and derive vegetation indicators such as Normalized Difference Vegetation Index (NDVI) (mean, min–max, and within-plot coefficient of variation), revealing fine-scale patterns in crop vigor and heterogeneity that are obscured at coarser resolutions. We then contrast these plot-level metrics with Sentinel-2 and Landsat-8 views to illustrate mixed-pixel effects and the implications for area estimation, monitoring, and operational decision-making.
By the end of this chapter, you will be able to:
Understand the UAV imagery workflow from flight planning to mapping output
Load and visualize RGB (true color) and NDVI UAV imagery in R
Load and manage vector data in R
Extract plot-level zonal statistics from UAV imagery
Compare spatial resolutions of UAV versus publicly available satellite imagery
The study area comprises a single census enumeration area located on the northeastern coast of Rarotonga island in the Cook Islands encompassing 17.6 hectares. This area was selected to enable cross-validation with area estimates from the 2022 Agriculture Census and post-Agriculture Census enumeration survey. The terrain in the study area is characterized by flat topography extending from inland residential areas to the beach and ocean along the eastern boundary. Land use is mixed, including residential, sports facilities (i.e., rugby field), and fragmented agricultural plots scattered throughout the area.
UAV imagery of the study area was captured on 28 August 2023 at midday. Two flight missions were conducted: the first using an RGB camera captured 347 aerial images (Figure 1), while the second using a multispectral sensor captured 318 images. Flight missions were planned using DJI Flight Planner software, where operators specified resolution requirements by setting flight height (~100 meters) and image overlap percentages (~80 percent) to ensure adequate coverage.